Graphic Processing Unit-accelerated optimization of hydropower

Graphic processing units (GPUs) have seen an explosive growth in use and performance over the last decade. Industries for gaming, cryptocurrency and artificial intelligence are all powered by these processors. The GoHydro project will pave the way for use of this computational power in optimal use of natural resources in the form of hydropower. Hydropower is the backbone of the Norwegian power system, and a key enabler for integration of new renewable sources like solar and wind. The energy system is going through a new phase due to climate changes, new intermittent production and political uncertainty. The number of scenarios needed to capture potential future development is larger than ever. This requires the next generation of hydropower optimization tools to handle huge amounts of data. The GPU architecture is perfectly suited for this, and the goal in GoHydro is to provide robust decisions based on a greater awareness of possible scenarios than the tools used today. There are ongoing large investments in capacity in the Norwegian hydropower system. The framework GoHydro will be built on is aiming at capturing both long-term strategic energy storage and short-term local storage in an optimal way for the energy system. This is a novel task that requires new computational power which will be provided by the GPUs.

In GoHydro we aim for a novel algorithm that can be used both for short and long-term planning horizons, and where the level of detail and the aspects that shall be considered can be chosen by the user. In this way the user only needs to be familiar with one model, and by adjusting the input data and the planning horizon, the user can tailor each run to her or his own needs. To facilitate such an algorithm and at the same time achieve high computational performance, GoHydro will utilize the immense computing power of modern graphics processing units (GPUs). We plan on building on a GPU-based proof-of-concept algorithm that the project consortium has prototyped within the last 2.5 years for modelling the water values over a long planning period. To achieve this, the following secondary objectives must be reached. - Adding more details to the proof-of-concept to enable more rich models that are needed for short-term planning horizons and enable long-term planning with more details. - Improve efficiency both by algorithmic and GPU programming techniques - Include uncertainty in the model - Provide feasible solutions. The current proof-of-concept provides a solution that is close to the optimal feasible solution but is in itself infeasible. - Take discrete decisions into account. A hydropower generator has certain constraints that need to be fulfilled for it to be able to run. Hence a feasible short-term schedule needs to include the decisions which units to run, and which to not - Provide theoretical foundation for the algorithm. Discuss the properties, conditions, and limitations of the resulting method. - Dissemination